Real-time classification of events in high energy physics is essential to deal with huge amounts of data, produced by proton-proton collisions in ATLAS detector at Large Hadron Collider in CERN. With this work we have implemented a triggering mechanism method for saving relevant data, based on machine learning. In comparison with the state of the art machine learning methods (gradient boosting and deep neural networks) shortcomings of Support Vector Machines (SVM) have been compensated with extensive feature engineering. Method has been evaluated with special metrics (average median significance) suggested by the domain experts. Our method achieves significantly higher precision and 8% lower average median significance than the current state of the art method used at ATLAS detector (XGBoost).